Aggressive release consistency and efficient lock implementation for software distributed shared memory
Guardado en:
| Publicado en: | ProQuest Dissertations and Theses (1997) |
|---|---|
| Autor principal: | |
| Publicado: |
ProQuest Dissertations & Theses
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | Distributed shared memory (DSM) allows processes to view the physically distributed memory as a globally shared virtual memory, possibly overcoming major obstacles of the widespread use of distributed-memory multiprocessors, while retaining the attractive features of low cost and good scalability common to distributed-memory machines. Lazy release consistency (LRC) (6) is proved to be most efficient among known software-based DSM models proposed. As a software-based DSM system is especially sensitive to the traffic amount over the network, LRC is preferred because it efficiently reduces traffic by postponing data coherence until the time of acquire such that data coherence messages are not sent over the network unless they are explicitly demanded by the next processor which fetches the shared data. In this thesis, we propose a new software DSM model which further postpones the enforcement of data coherence until the time of the first shared memory access after the acquire, leading to an aggressive implementation of release consistency and thus more reduction in the number of messages transferred over the network than LRC. This proposed model is referred to as aggressive release consistency, which is particularly suitable for the software DSM implementation to achieve faster program execution. The introduced model has been developed and evaluated on an Ethernet-connected network of workstations using various benchmark programs. The experimental results indicate that the proposed scheme leads to better performance than the most efficient DSM implementation known so far. DSM relies heavily on lock synchronization to maintain data coherency. For a large system size, traffic congestion caused by synchronization degrades system performance significantly. To investigate the synchronization issue in DSM, both shared memory system and distributed memory system are examined in regard to lock synchronization. In this thesis, we introduce an efficient lock scheme for the shared memory system, and the scheme keeps synchronization traffic low and avoids serious hot-spot contention. Extensive simulation of the proposed scheme was conducted and the lower bound on the elapsed time was derived. Our simulation results demonstrate that the proposed lock scheme indeed achieves better performance than prior techniques, with its elapsed time close to the lower bound for the whole range of simulated system sizes, thus promising good scalability for large systems. For the distributed memory system, we evaluated various distributed lock schemes on the IBM SP2 machine. The empirical results are compared in terms of such criteria as the number of message exchanges and the response time. The results indicate that, when every processor enters the critical section only once before encountering a barrier, the improved Ring algorithm (4) is found to outperform others under a heavy load; but the Star algorithm and the CSL algorithm (3) prevail when the request rate becomes light. Since the number of requests (to the critical section) a site generates before encountering a barrier dictates the degree of contention, the best solution to mutual exclusion in distributed memory systems is determined by how participating sites generate their mutual exclusion requests. |
|---|---|
| ISBN: | 9780591361438 |
| Fuente: | ProQuest Dissertations & Theses Global |